GPT-4oã®ä¸å›½èªžãƒˆãƒ¼ã‚¯ãƒ³ã¯ãƒãƒ«ãƒŽã¨ã‚¹ãƒ‘ムã«æ±šæŸ“ã•ã‚Œã¦ã„ã‚‹2024å¹´5月ã«OpenAIã®æœ€æ–°AIモデル「GPT-4oã€ãŒç™ºè¡¨ã•ã‚Œã¾ã—ãŸã€‚テã‚ストや音声ã€ã‚«ãƒ¡ãƒ©ã®å…¥åŠ›ã‚’人間ã¨åŒã˜ãらã„ã®é€Ÿåº¦ã§å‡¦ç†å¯èƒ½ã¨ã„ã†é«˜ã„性能ã§ã‚ã‚‹ã“ã¨ãŒå ±ã˜ã‚‰ã‚Œã¦ã„ã‚‹ã®ã§ã™ãŒã€ä¸€æ–¹ã§ã€ä¸å›½èªžãƒ¦ãƒ¼ã‚¶ãƒ¼ã‹ã‚‰ã¯ã€ãƒˆãƒ¬ãƒ¼ãƒ‹ãƒ³ã‚°ã«å¤§ããªå•é¡Œã‚’抱ãˆã¦ã„ã¦ã€ãƒˆãƒ¼ã‚¯ãƒ³ãƒ‡ãƒ¼ã‚¿ãŒæ±šæŸ“ã•ã‚Œã¦ã„ã‚‹ã“ã¨ãŒæŒ‡æ‘˜ã•ã‚Œã¦ã„ã¾ã™ã€‚ Just wrote a script to further investigate how the corpus used to train the gpt4o tokenizer is polluted by Internet scams. The results are quite interesting... 🤦â€â™‚ï¸ðŸ¤¦â€â™‚ï¸ðŸ¤¦â€â™‚ï¸https://t.co/Fc2T4rSHix https://t.co/Q1Syh9amJn pic.twitter.com/lQ1u
「リートンã€ã«è„†å¼±æ€§ ～データベースã‹ã‚‰ãƒ¦ãƒ¼ã‚¶ãƒ¼æƒ…å ±ã‚’çªƒå–å¯èƒ½ãªå•é¡Œã‚’ä¿®æ£ï¼ç¾æ™‚点ã§ã¯ãƒ‡ãƒ¼ã‚¿ã‚’悪用ã•ã‚ŒãŸå ±å‘Šã¯ãªã—
ChatGPT API ã§åŒºåˆ‡ã‚Šç·šã‚’表ç¾ã™ã‚‹æœ€é©ãªæ–‡å—列を見ã¤ã‘ãŸä»¶
GPT-1→GPT-2→GPT-3→GPT-3.5→ChatGPT→GPT-4ã¾ã§ã®é€²åŒ–ã®è»Œè·¡ã¨é•ã„ã‚’ã¾ã¨ã‚ã¦ã¿ãŸï½œã‚¹ã‚¿ãƒ“ジ
{{#tags}}- {{label}}
{{/tags}}